1. Aprendemos de los conjuntos de datos

• Raschka, S., Liu, Y, Mirjalili, V. (2022). Machine Learning with PyTorch and Scikit-Learn. Develop machine learning and deep learning models with Python. Packt Publishing. Birmingham, UK. Pag. 3. Making predictions about the future with supervised learning.

• Kneusel, R.T. (2021). Practical Deep Learning - A Python-Based Introduction. First Edition. No Starch Press, Inc. San Francisco, USA. Pag. 69. Training, Validation, and Test Data - The Three Subsets.

• Zheng, A., Casari, A. (2018). Feature Engineering for Machine Learning. First Edition. O’Reilly Media, Inc. Sebastopol, CA, USA. Pag. 4. The Machine Learning Pipeline.

1.1. Seleccionamos métodos y herramientas

• Label-Studio

• Label-Studio es una herramienta que utilizamos para etiquetar datos. Permite entre otras cosas, el trabajo conjunto en tiempo real sobre imágenes, audios, textos y vídeos.

• Label-Studio

• Por ejemplo, podríamos analizar la siguiente observación acústica real:

from IPython.display import Audio

wav_dir = '/home/kasparov/Documentos/20190114_vpoblete/20221023_aumilab/JupyterBookUACh/audios/'

audio0 = wav_dir + "siren.wav"
Audio(audio0)

1.2. Programamos en lenguaje Python y usamos de librerías

• Observemos audios reales:

audio1 = wav_dir + "alarm.wav"
Audio(audio1)

audio2 = wav_dir + "bird.wav"
Audio(audio2)

audio3 = wav_dir + "dog.wav"
Audio(audio3)

audio4 = wav_dir + "helicopter.wav"
Audio(audio4)

audio5 = wav_dir + "motorcycle.wav"
Audio(audio5)

audio6 = wav_dir + "music.wav"
Audio(audio6)

audio7 = wav_dir + "shout.wav"
Audio(audio7)

audio8 = wav_dir + "speech.wav"
Audio(audio8)

class Power:
    
    def __init__(self, name, wav_dir):
        self.name = name
        self.kind = ''
        self.duration = 0.0
        self.power = 0.0
        self.calculator(name, wav_dir)
        
    def calculator(self, audio, wav_dir):
        self.kind_set(audio)
        self.audio_average_power(wav_dir)
    
    def audio_average_power(self, wav_dir):
        
        wav_path = os.path.join(wav_dir, self.name)
        waveform, sample_rate = torchaudio.load(wav_path)
        num_channels, num_frames = waveform.shape
        
        if waveform.numpy().shape[1] != 0:
            
            ft_wf = np.fft.fft(waveform.numpy())/(math.sqrt(waveform.numpy().shape[1]))
            
            parseval_waveform = np.sum(waveform.numpy()**2)
            parseval_fourier_transform = np.sum(np.abs(ft_wf)**2)
            
            if (np.float32(round(parseval_waveform,1)) == np.float32(round(parseval_fourier_transform,1))):
                
                self.power = parseval_waveform
                
                duration = torch.arange(0, num_frames) / sample_rate
                self.duration = duration.numpy()[-1]
    
    def kind_set(self, audio):
        text = audio.split('.')
        self.kind = text[0]

import os
import torch
import torchaudio
import numpy as np
import math
import pandas as pd
import matplotlib.pyplot as plot

audio_class = []
audio_class.append(Power('alarm.wav', wav_dir))
audio_class.append(Power('bird.wav', wav_dir))
audio_class.append(Power('dog.wav', wav_dir))
audio_class.append(Power('helicopter.wav', wav_dir))
audio_class.append(Power('motorcycle.wav', wav_dir))
audio_class.append(Power('music.wav', wav_dir))
audio_class.append(Power('rain.wav', wav_dir))
audio_class.append(Power('shout.wav', wav_dir))
audio_class.append(Power('speech.wav', wav_dir))
audio_class.append(Power('wind.wav', wav_dir))

1.2.1. Tabla

data = []
for i in range(len(audio_class)):
    
    data.append([audio_class[i].name, audio_class[i].kind, audio_class[i].duration, audio_class[i].power])

df = pd.DataFrame(data, columns = ['file name', 'event', 'duration', 'power'])
pd.set_option('display.float_format', '{:.0f}'.format)
df

	file name	event	duration	power
0	alarm.wav	alarm	60	36324
1	bird.wav	bird	31	133
2	dog.wav	dog	6	5737
3	helicopter.wav	helicopter	20	50644
4	motorcycle.wav	motorcycle	20	8359
5	music.wav	music	30	31215
6	rain.wav	rain	28	872
7	shout.wav	shout	10	9410
8	speech.wav	speech	15	4239
9	wind.wav	wind	38	15681

1.2.2. Gráfico

ax = df.plot.scatter(x='duration', y='event', c='power', colormap='jet', figsize=(5,4))